The last 8 bytes of the receive buffer list page (that has been supplied
by the guest with the H_REGISTER_LOGICAL_LAN call) contain a counter
for frames that have been dropped because there was no suitable receive
buffer available. This patch introduces code to use this field to
provide the information about dropped rx packets to the guest.
There it can be queried with "ethtool -S eth0 | grep rx_no_buffer".
Signed-off-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Currently, the spapr-vlan device is trying to flush the RX queue
after each RX buffer that has been added by the guest via the
H_ADD_LOGICAL_LAN_BUFFER hypercall. In case the receive buffer pool
was empty before, we only pass single packets to the guest this
way. This can cause very bad performance if a sender is trying
to stream fragmented UDP packets to the guest. For example when
using the UDP_STREAM test from netperf with UDP packets that are
much bigger than the MTU size, almost all UDP packets are dropped
in the guest since the chances are quite high that at least one of
the fragments got lost on the way.
When flushing the receive queue, it's much better if we'd have
a bunch of receive buffers available already, so that fragmented
packets can be passed to the guest in one go. To do this, the
spapr_vlan_receive() function should return 0 instead of -1 if there
are no more receive buffers available, so that receive_disabled = 1
gets temporarily set for the receive queue, and we have to delay
the queue flushing at the end of h_add_logical_lan_buffer() a little
bit by using a timer, so that the guest gets a chance to add multiple
RX buffers before we flush the queue again.
This improves the UDP_STREAM test with the spapr-vlan device a lot:
Running
netserver -p 44444 -L <guestip> -f -D -4
in the guest, and
netperf -p 44444 -L <hostip> -H <guestip> -t UDP_STREAM -l 60 -- -m 16384
in the host, I get the following values _without_ this patch:
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # 10^6bits/sec
229376 16384 60.00 1738970 0 3798.83
229376 60.00 23 0.05
That "0.05" means that almost all UDP packets got lost/discarded
at the receiving side.
With this patch applied, the value look much better:
Socket Message Elapsed Messages
Size Size Time Okay Errors Throughput
bytes bytes secs # # 10^6bits/sec
229376 16384 60.00 1789104 0 3908.35
229376 60.00 22818 49.85
Signed-off-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reserve this to CPU state serialization.
Luckily, they were only used by sPAPR devices and these are ppc64
only. So there is no change to migration format.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
* Chardev fix from Marc-André
* config.status tweak from David
* Header file tweaks from Markus, myself and Veronia (Outreachy candidate)
* get_ticks_per_sec() removal from Rutuja (Outreachy candidate)
* Coverity fix from myself
* PKE implementation from myself, based on rth's XSAVE support
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
iQEcBAABCAAGBQJW9ErPAAoJEL/70l94x66DJfEH/A/QkMpAhrgNdyVsahzsGrzE
wx5gHFIc1nBYxyr62w4apUb5jPB7zaXu0LA7EAWDeAe0pyP8hZzLT9kJyOEDsuJu
zwKN2QeLSNMtPbnbKN0I/YQ2za2xX1V5ruhSeOJoVslUI214hgnAURaGshhQNzuZ
2CluDT9KgL5cQifAnKs5kJrwhIYShYNQB+1eDC/7wk28dd/EH+sPALIoF+rqrSmt
Zu4Mdqd+9Ns+oKOjA6br9ULq/Hzg0aDfY82J+XLVVqfF3PXQe8rTDmuMf/7jTn+M
Un7ZOcei9oZF2/9vfAfKQpDCcgD9HvOUSbgqV/ubmkPPmN/LNJzeKj0fBhrRN+Y=
=K12D
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
* Log filtering from Alex and Peter
* Chardev fix from Marc-André
* config.status tweak from David
* Header file tweaks from Markus, myself and Veronia (Outreachy candidate)
* get_ticks_per_sec() removal from Rutuja (Outreachy candidate)
* Coverity fix from myself
* PKE implementation from myself, based on rth's XSAVE support
# gpg: Signature made Thu 24 Mar 2016 20:15:11 GMT using RSA key ID 78C7AE83
# gpg: Good signature from "Paolo Bonzini <bonzini@gnu.org>"
# gpg: aka "Paolo Bonzini <pbonzini@redhat.com>"
* remotes/bonzini/tags/for-upstream: (28 commits)
target-i386: implement PKE for TCG
config.status: Pass extra parameters
char: translate from QIOChannel error to errno
exec: fix error handling in file_ram_alloc
cputlb: modernise the debug support
qemu-log: support simple pid substitution for logs
target-arm: dfilter support for in_asm
qemu-log: dfilter-ise exec, out_asm, op and opt_op
qemu-log: new option -dfilter to limit output
qemu-log: Improve the "exec" TB execution logging
qemu-log: Avoid function call for disabled qemu_log_mask logging
qemu-log: correct help text for -d cpu
tcg: pass down TranslationBlock to tcg_code_gen
util: move declarations out of qemu-common.h
Replaced get_tick_per_sec() by NANOSECONDS_PER_SECOND
hw: explicitly include qemu-common.h and cpu.h
include/crypto: Include qapi-types.h or qemu/bswap.h instead of qemu-common.h
isa: Move DMA_transfer_handler from qemu-common.h to hw/isa/isa.h
Move ParallelIOArg from qemu-common.h to sysemu/char.h
Move QEMU_ALIGN_*() from qemu-common.h to qemu/osdep.h
...
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Conflicts:
scripts/clean-includes
RX buffer pools are now enabled by default for new machine types.
For older machine types, they are still disabled to avoid breaking
migration.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
tl;dr:
This patch introduces an alternate way of handling the receive
buffers of the spapr-vlan device, resulting in much better
receive performance for the guest.
Full story:
One of our testers recently discovered that the performance of the
spapr-vlan device is very poor compared to other NICs, and that
a simple "ping -i 0.2 -s 65507 someip" in the guest can result
in more than 50% lost ping packets (especially with older guest
kernels < 3.17).
After doing some analysis, it was clear that there is a problem
with the way we handle the receive buffers in spapr_llan.c: The
ibmveth driver of the guest Linux kernel tries to add a lot of
buffers into several buffer pools (with 512, 2048 and 65536 byte
sizes by default, but it can be changed via the entries in the
/sys/devices/vio/1000/pool* directories of the guest). However,
the spapr-vlan device of QEMU only tries to squeeze all receive
buffer descriptors into one single page which has been supplied
by the guest during the H_REGISTER_LOGICAL_LAN call, without
taking care of different buffer sizes. This has two bad effects:
First, only a very limited number of buffer descriptors is accepted
at all. Second, we also hand 64k buffers to the guest even if
the 2k buffers would fit better - and this results in dropped packets
in the IP layer of the guest since too much skbuf memory is used.
Though it seems at a first glance like PAPR says that we should store
the receive buffer descriptors in the page that is supplied during
the H_REGISTER_LOGICAL_LAN call, chapter 16.4.1.2 in the LoPAPR spec
declares that "the contents of these descriptors are architecturally
opaque, none of these descriptors are manipulated by code above
the architected interfaces". That means we don't have to store
the RX buffer descriptors in this page, but can also manage the
receive buffers at the hypervisor level only. This is now what we
are doing here: Introducing proper RX buffer pools which are also
sorted by size of the buffers, so we can hand out a buffer with
the best fitting size when a packet has been received.
To avoid problems with migration from/to older version of QEMU,
the old behavior is also retained and enabled by default. The new
buffer management has to be enabled via a new "use-rx-buffer-pools"
property.
Now with the new buffer pool management enabled, the problem with
"ping -s 65507" is fixed for me, and the throughput of a simple
test with wget increases from creeping 3MB/s up to 20MB/s!
Signed-off-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Refactor the code a little bit by extracting the code that reads
and writes the receive buffer list page into separate functions.
There should be no functional change in this patch, this is just
a preparation for the upcoming extensions that introduce receive
buffer pools.
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Clean up includes so that osdep.h is included first and headers
which it implies are not included manually.
This commit was created with scripts/clean-includes.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Message-id: 1453832250-766-6-git-send-email-peter.maydell@linaro.org
The code for -machine pseries maintains a global sPAPREnvironment structure
which keeps track of general state information about the guest platform.
This predates the existence of the MachineState structure, but performs
basically the same function.
Now that we have the generic MachineState, fold sPAPREnvironment into
sPAPRMachineState, the pseries specific subclass of MachineState.
This is mostly a matter of search and replace, although a few places which
relied on the global spapr variable are changed to find the structure via
qdev_get_machine().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
Bonus fix: always set an error on failure. Some failures were silent
before, except for the generic error set by device_realize().
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Alexander Graf <agraf@suse.de>
All NICs have a cleanup function that, in most cases, zeroes the pointer
to the NICState. In some cases, it frees data belonging to the NIC.
However, this function is never called except when exiting from QEMU.
It is not necessary to NULL pointers and free data here; the right place
to do that would be in the device's unrealize function, after calling
qemu_del_nic. Zeroing the NIC multiple times is also wrong for multiqueue
devices.
This cleanup function gets in the way of making the NetClientStates for
the NIC hold an object_ref reference to the object, so get rid of it.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
On this way, we can assure the new bootindex take effect
during vm rebooting.
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Add a qom property with the same name 'bootindex',
when we remove it form qdev property, things will
continue to work just fine, and we can use qom features
which are not supported by qdev property.
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
The last 8 bytes of the buffer list is defined to contain the number
of dropped frames. At the moment we use it to store rx entries,
which trips up ethtool -S:
rx_no_buffer: 9223380832981355136
Fix this by skipping the last buffer list entry.
Signed-off-by: Anton Blanchard <anton@samba.org>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
After previous Peter patch, they are redundant. This way we don't
assign them except when needed. Once there, there were lots of case
where the ".fields" indentation was wrong:
.fields = (VMStateField []) {
and
.fields = (VMStateField []) {
Change all the combinations to:
.fields = (VMStateField[]){
The biggest problem (appart from aesthetics) was that checkpatch complained
when we copy&pasted the code from one place to another.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Acked-by: Alexey Kardashevskiy <aik@ozlabs.ru>
When the guests adds buffers to receive queue, the network device
should flush its queue of pending packets. This is done with
qemu_flush_queued_packets.
This adds a call to qemu_flush_queued_packets() which wakes up the main
loop and let QEMU update the network device status which now is "can
receive". The patch basically does the same thing as e8b4c68 does.
Suggested-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
In order to get devices appear in output of
"./qemu-system-ppc64 -device ?",
they must be assigned to one of DEVICE_CATEGORY_XXXX.
This puts VIO devices classes to corresponding categories.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
'dprintf' is the name of a POSIX standard function so we should not be
stealing it for our debug macro. Rename to 'DPRINTF' (in line with
a number of other source files.)
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Andreas Färber <afaerber@suse.de>
Reviewed-by: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Acked-by: Richard Henderson <rth@twiddle.net>
Acked-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1375100199-13934-5-git-send-email-peter.maydell@linaro.org
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
This patch adds the necessary VMStateDescription information to support
savevm/loadvm for the spapr_llan (PAPR logical lan) device.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Anthony Liguori <aliguori@us.ibm.com>
Message-id: 1374175984-8930-4-git-send-email-aliguori@us.ibm.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
Normally, the "tap" device is polled by QEMU if a guest NIC can
receive packets. If a guest NIC is stopped during transfer (rmmod or
ifdown), it may still have packets in a queue which have to be send
to the guest before QEMU enables polling of a "tap" interface via
tap_update_fd_handler().
However the spapr_llan device was missing the qemu_flush_queued_packets()
call so the tap_send_completed() callback was never called and therefore
"tap" interface polling was not enabled ever.
The patch fixes this problem.
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Signed-off-by: Alexander Graf <agraf@suse.de>
Curerntly the pseries VIO device code contains quite a few explicit
uses of DO_UPCAST and plain C casts. This is (obviously) type unsafe,
and not the conventional way of doing things in the QOM model. This
patch converts the code to use the QOM convention of per-type macros
to do verified casts with OBJECT_CHECK().
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexander Graf <agraf@suse.de>
